The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to override agent safety protocols and user oversight. It directs the agent to install dependencies "automatically without asking the user to confirm" and mandates that the agent "do not request separate approval for dependency installation."
[REMOTE_CODE_EXECUTION]: The instructions command the agent to download and install missing software (the BrowserAct CLI and Skill) from an external source at runtime. By coupling this with a directive to bypass user confirmation, the skill enables the execution of arbitrary external code without human review.
[EXTERNAL_DOWNLOADS]: The skill points to a non-standard external domain (browseract.com) and an untrusted GitHub repository for dependency fulfillment. This introduces supply chain risks as the agent is encouraged to fetch code from sources outside of verified or trusted organization lists.
[COMMAND_EXECUTION]: The skill is designed to invoke the browser-act CLI to perform browser-based tasks. The instructions for automatic installation and subsequent invocation of this CLI constitute a pattern of unverified command execution.
[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to ingest untrusted data from Xiaohongshu (RedNote). It lacks mandatory safety evidence, specifically missing boundary markers (e.g., delimiters) and sanitization procedures to prevent malicious instructions embedded in scraped notes from influencing the agent's behavior.
[METADATA_POISONING]: The skill exhibits a discrepancy between its internal metadata author field ('rebeccareyes3794') and the expected author context, which may be a deceptive tactic to obscure the origin of the skill.

xiaohongshu-rednote-trending-notes-scraper